This action might not be possible to undo. Are you sure you want to continue?
News-Sharing Communities in Twitter
Ama¸c Herda˘gdelen, Wenyun Zuo, Alexander Gard-Murray and Yaneer Bar-Yam
New England Complex Systems Institute
238 Main Street Suite 319, Cambridge, Massachusetts 02142, US
(Dated: February 17, 2012)
The importance of collective social action in current events is manifest in the Arab Spring and
Occupy movements. Electronic social media have become a pervasive channel for social interac-
tions, and a basis of collective social response to information. The study of social media can reveal
how individual actions combine to become the collective dynamics of society. Characterizing the
groups that form spontaneously may reveal both how individuals self-identify and how they will
act together. Here we map the social, political, and geographical properties of news-sharing com-
munities on Twitter, a popular micro-blogging platform. We track user-generated messages that
contain links to New York Times online articles and we label users according to the topic of the
links they share, their geographic location, and their self-descriptive keywords. When users are
clustered based on who follows whom in Twitter, we ﬁnd social groups separate by whether they
are interested in local (NY), national (US) or global (cosmopolitan) issues. The national group
subdivides into liberal, conservative and other, the latter being a diverse but mostly business ori-
ented group with sports, arts and other splinters. The national political groups are based across
the US but are distinct from the national group that is broadly interested in a variety of topics. A
person who is cosmopolitan associates with others who are cosmopolitan, and a US liberal / con-
servative associates with others who are US liberal / conservative, creating separated social groups
with those identities. The existence of “citizens” of local, national and cosmopolitan communities
is a basis for dialog and action at each of these levels of societal organization.
Who interacts with whom aﬀects the characteristic behavior of social systems. The
rapid development of electronic social media  may radically change both the structure
of social networks and the actions they take [2, 3]. Here we map the structure of news
sharing social groups (Figure 1), revealing that they are organized around local, national
and cosmopolitan concerns. The national interest group is partitioned into three subgroups:
two that are politically liberal and conservative, and a third that is interested in a variety
of topics, including business, sports and arts.
Social links determine what kind of information people are exposed to, and thus how they
respond to ongoing events, especially when time is a critical factor in the response. What
information is selected for transmission also inﬂuences the development of perspectives.
Strongly linked groups that rapidly share information may react together and with shared
views [4, 5, 6]. The wider social propagation and inﬂuence of information is then limited by
the strength of “weak ties” between these groups . An understanding of society not only
depends on characterizing the behavior of individuals and their immediate links, but also
on identifying emergent social collectives. For example, individuals might group by political
orientation, economic class, race, family, profession, age, religion or geography, or by domain
FIG. 1: News sharing network. Nodes are individuals who predominantly share news stories on top-
ics given by the legend. Links are “follow” relationships between individuals. Cosmopolitan, local
scene, national liberal, national conservative, and national diverse are tightly connected groups.
of interest such as business, sports, arts and politics.
One of the universal properties of complex systems is substructure [8, 9], the charac-
teristics of which are essential to a particular system’s distinct qualities . The role of
substructure is apparent in an analysis of the information ﬂows in the human brain, where
sensory, motor and other functional separations of information are key to the overall infor-
mation processing. Similarly, we can expect that the diﬀerentiation across social groups will
play an important role in the behaviors and function of society.
Communication media, both peer-to-peer and broadcast, connect individuals separated
in space and time. The patterns of communication change not only the experience of an
individual but also the structure and behavior of our social systems.
Before electronic communication, strongly linked social groups were primarily geographic
[2, 3, 10, 11]. Individual and group identities were deﬁned by the frequent interactions
of each individual with other members of the group. The variations among groups, their
distinct cultures, arose from variations in local conditions and historical events. Individuals
might change their associations by moving, and groups might change their location by
migrating. As a way of understanding the changing nature of human social organization,
we might calibrate it to the growing frequency of longer range interactions, and the extent
to which communication is one-to-one (peer-to-peer) or one-to-many, the latter being a
characteristic of mass media. From messengers, to letters, telegrams, telephone, e-mail and
e-messaging, peer-to-peer communications became easier and more frequent. From public
presentations, to printing presses, radio, cinema, and television, mass media expanded the
scope of recipients and restricted the sources of communication. The internet has merged
peer-to-peer and broadcast media, increasing the ﬂexibility of both in the form of social
As the ease of long range communication increases, the dominant role of geographical
association may diminish, and the process of group formation becomes motivated by acts of
selection—selection by individuals of those with whom they choose to communicate. Social
associations can be based upon any criteria. Released from the constraints of geography,
people may choose to associate with others based upon what they consider most important
and their self-identity. Instead of a local or national identity, individuals may choose a
wide variety of other forms of identity. A person focused on music could choose only to
communicate with others who are focused on music. Religiously focused individuals may
communicate solely with others with shared religious views. Individuals who primarily value
family ties may focus on those relationships. Moreover, the choices of association reinforce
the form of identity that led to those associations.
Individual choices of who to interact with result in the overall communication structure
of society. The communication structure is then responsible for the collective behaviors
that arise when news about events, and opinion formation, result in social movements and
collective decision making [12, 13, 14]. Understanding this structure and dynamics can be
fundamental to our ability to characterize society. The availability of detailed information
about social media provides a new way to advance our understanding of the dynamics
of individual choices of association and social organization. Our objective is to advance
this understanding by starting to characterize the spontaneous development of collective
structures in social media. We focus speciﬁcally on those structures associated with news
information sharing, which are the most directly relevant to societal response to news events.
Hundreds of millions of people around the world use social media and micro-blogging
platforms as a real-time information source [15, 16, 17]. Here we identify news-sharing com-
munities in the micro-blogging platform Twitter. Twitter users post very short (at most
140-character) posts called tweets. More than a hundred million tweets are posted every
day . These tweets are typically about what users do, think, or experience and want to
share with other people . Sharing URLs and reported news constitutes a signiﬁcant frac-
tion of user activities in Twitter [20, 21]. The brevity of tweets and their rapid distribution
facilitates sharing and receiving the news as events unfold .
Twitter enables individuals general access to tweets, but also allows people to construct
social connections that form primary channels of communication. Twitter users identify
other users to “follow” in order to be notiﬁed of their tweets. Each user can be thought of
as a node in a network, and the relationships as links between them.
A “follow relation” in Twitter may reﬂect various kinds of relationships. Preexisting
friends may follow each other, but there are also ways to discover others to follow within the
Twitter system. Since users share what is important to them, the way the Twitter society
organizes itself both reﬂects and inﬂuences social communication. Research has conﬁrmed
that Twitter users with similar interests tend to connect to each other . Analyses suggest
that Twitter serves more for information sharing than for casual conversation [15, 23]. There
is also research that has associated speciﬁc Twitter users or user categories (celebrities, me-
dia, organizations, and blog writers) with speciﬁc intentions such as information providing,
information seeking and friendship/conversation [15, 19]. For example, celebrities show an
elevated interest in sport-related news, media accounts are more interested in US-related
news, and organizations show more interest in science, technology and world news .
The ability to track the social relationships and the posted content in social media enables
studies of the communities that are forming and their emerging collective behaviors. Here
we study the overall structure of certain communities that have spontaneously formed in
Twitter and we characterize their social, political, and geographic natures. The topology
of a network has an important eﬀect on dynamics of the network  and its response to
external perturbations , as can be seen in epidemics  and information spread in social
networks . However, our focus is on the characteristics of the spontaneous social groups
that are formed and the insights that this gives about user self-identiﬁcation, social priorities
and collective social response.
We identify social groups formed out of users who tweet or retweet New York Times
online articles. The New York Times is a major traditional news media outlet that is read
by millions of people all over the world. We use three dimensions to characterize users: topics
of interest, geographical locations, and self-description. The topics of interest are based on
the articles that the users share. The topic category is identiﬁable in the article’s URL, i.e.
www.nytimes.com/2011/10/05/science/... is a science related article. Geographical location
is identiﬁed by analyzing the location ﬁeld of user proﬁles in Twitter. Self-description is
based on analyzing the terms the users use to describe themselves in their Twitter proﬁle
“bio” ﬁeld. The multiple dimensions of identity provide insights into the nature of the groups
that have spontaneously formed in the Twitter network. Despite its New York focus, the
global reach of the New York Times enables us to consider the implications of the results
for developing national and global communities. Details of the analysis are given in the
We collected tweets containing a New York Times article link over a ﬁfteen day period,
September 14 - 29, 2011. This resulted in a collection of 521,733 tweets posted by 223,950
unique users. We considered only users who posted at least three URLs from the same
category. We also obtained the “following” relationships (who follows whom) among these
users, and constructed a graph of the users and their follow relations. The largest connected
component of the graph contains 8,106 nodes and 163,850 links. Each user was labeled with
the topic that s/he posted the most, with ties broken in favor of less popular topics. Details
are provided in Appendix A.
The news sharer network we obtained is shown in Figure 2. The layout, shown at the
top, is completely determined by the connections between users. Users who are connected
to each other by a follow relationship are pulled together by virtual springs and those who
are not are pushed apart . The visual layout then naturally identiﬁes those nodes that
belong to highly connected groups. No information about the identity or interests of the
users is used to create the layout. The clustering of the network manifests a strong social
separation that can also be shown using a clustering algorithm that identiﬁes the left and
right parts of the network and further divides each to upper and lower clusters. Details of
the layout generation and cluster generation are provided in Appendix B.
In the main image in Figure 2, the user nodes are colored according to the article topics
that users most often share in tweets. Topic colors are not randomly scattered. In some
but not all cases the same colored users are clustered together. The cases where they do are
consistent with the idea that users with the same interests tend to follow each other. For
instance, users who are interested in the topic “world” are clustered in the upper-left part of
the network, and users who are interested in the topic “US” are clustered in the upper-right
The most prevalent topics of interest for the primary clusters are shown in Figure 3.
The division of left (A) and right (B) clusters most prominently separates those who are
interested in world news (cluster A) from those interested in US-based news (cluster B). We
further divide cluster A into upper and lower sub-clusters and ﬁnd that the dominant topic
in upper A is world news whereas lower A has a more diverse set of topics with business,
US, arts, fashion, and sports being well represented in this sub-cluster.
The results show that the clusters are quite distinct and begin to reveal their nature. For
example, it is not the case that each news topic has its own cluster. Instead, some topics
diﬀerentiate the clusters, while others are grouped together. The clear separation between
US and world news suggests that geography might be the key to the structure of the social
network (A geographical location analysis is given in Appendix C). However, the sense in
which geography plays a role is through the scope of interest: local, national and global.
Identifying the most distinctive aspects of each cluster in all three dimensions (topic,
location, and biography) provides a more complete picture as shown in Figure 4 (Details are
FIG. 2: (top) Network of Twitter users who share New York Times online articles. Links are
“follow” relationships between the users. The network layout is obtained by pulling followers and
followees close to each other, while pushing apart unconnected nodes. The long range links that
are visible occur where relatively few links connect groups of nodes that are otherwise pushed far
apart. (main image) The same network layout with user nodes colored according to the topic of
articles they share most. Links are shaded with the color of the follower.
given in Appendix D).
• Upper A is formed of users interested in international topics, who live in various
cities around the world, including New York and Washington DC, are focused on
human rights and politics, and may themselves be journalists. We might call them
the cosmopolitan group. The small separate cluster in the upper left consists of users
FIG. 3: The same network structure given in Figure 2, links are omitted for clarity. For each of
the sub-communities (A, B, Upper A, Lower A, Upper B, and Lower B) the top three URL topic
distributions are identiﬁed and these categories are given in the bar charts. The width of each
bar-chart component is proportional to the number of users in the corresponding category. The
total width of each bar chart is proportional to the total number of users in each community.
giving their location as Bahrain, which appear due to the Arab Spring events during
the period of time we studied.
• Lower A is a New York oriented group, many of whom say they are located in New
York, interested in a diverse set of topics including world, US, business, arts, fashion
and sports. We can call them the local or “New York scene.”
• Upper B is interested primarily in US politics, is US based, with focal locations in
New York and Washington DC, and is predominantly liberal and progressive.
• Lower B, like Lower A, has a diverse set of topics with a stronger focus on business,
but also interested in world news, sports, fashion and the arts. It is geographically
1 0.5 0 0.5
0.5 0 0.5 1
0 0.5 0
0.5 0 0.5 0.5 0 0.5
0.5 0.5 0.5
upper A lower A upper A lower A upper B lower B upper B lower B
A B A B
FIG. 4: Comparison between clusters based on topic, location and biography attributes. Each
vertical axis shows the most dominant terms associated with that cluster. The top ﬁgure compares
Cluster A (left) and Cluster B (right), as well as the upper and lower domains shown by distinct
colors. The bottom left two axes and right two axes compare upper A to lower A, and upper B to
lower B, respectively.
spread across the US and has an apparently conservative bent. However, we found
that this orientation is speciﬁc to one subgroup of Lower B. Figure 2 shows that the
Lower B group itself separates into a top right closely coupled cluster and other less
strongly clumped ones. The top right group is primarily interested in US politics and
self-identiﬁes as conservative, and the rest of Lower B has a broad array of interests
and is not politically distinct.
Given the separation of the Lower B group into the conservative cluster and the rest,
we have ﬁve primary clusters that can be readily identiﬁed. A simple characterization of
these groups is that three are identiﬁed with New York, US and Global interests. The other
two clusters are also US based but are speciﬁcally liberal and conservative in their political
Aside from the main clusters, there are small topically focused clusters of users partic-
ularly in the lower part of Figure 1, that can be seen to be primarily interested in arts or
The results reveal intriguing insights about how the network is organized. Traditional
geographic groups would be associated with individual cities or countries. Freedom from
geography does not eliminate entirely geographical association but results in subject driven
communities that are interested in local (NY), national (US) and global (cosmopolitan) is-
sues. The national group subdivides into liberal and conservative political groups and a
diverse but mostly business oriented group with sports, arts and other splinters. A person
who is cosmopolitan associates with others who are cosmopolitan, and a US liberal / con-
servative associates with others who are US liberal / conservative, creating separated social
groups with those identities. Thus, while local associations are sometimes primary, individ-
uals may choose to associate with national or cosmopolitan groups. A signiﬁcant fraction of
the population have become so strongly identiﬁed with ideological camps that those iden-
tities drive their social associations. For those who are concerned about the polarization
of society into liberal and conservative camps [29, 30], the results have both positive and
negative connotations. The groups that stratify into local, national and cosmopolitan are
not politically polarized, but there are speciﬁc subgroups that are polarized into opposing
Our analysis is limited to New York Times news sharing and a speciﬁc window of time.
Individual Twitter users may represent individuals or institutions. Also, the existence of
speciﬁc keyword dominances is a prevalence measure and does not imply that all members
or even a majority of the members of a group have that property. Nevertheless, our results
manifest the strong clustering of social systems that results in speciﬁc group identity.
Recent research has focused on user types and network topologies in social media [15, 16,
19, 31, 32, 33, 34, 35]. Studies have considered the distance distribution of cell phone social
networks  and of a blog community . The latter manifests both short range and
long range connections, which our research suggests may be due to distinct sub-groups. Our
analysis also has implications for understanding the ability of individuals in social networks
to intentionally transmit messages globally [36, 38, 39]. This can be readily understood if
groups that are national and cosmopolitan in orientation exist and are even weakly known by
individuals of local groups. The widespread geography of members of cosmopolitan groups
enable them to broadly transmit messages.
In summary, social media, and in particular Twitter, have created new opportunities for
communication that are closely connected to daily lives of people, their demographics, and
social and political interests. The changing communication channels also promise signiﬁcant
changes in the dynamics of social response to events. Social media contribute to the ability
of groups to engage in activities, even including revolutions, that are underway around the
world . The availability of data about these social networks and communications also
provides new opportunities for understanding how social systems form and function. We
used a combination of text and network analysis to study how people share information, and
which patterns emerge from the social interactions and interests. The analysis reveals the
natural clustering of society. We ﬁnd individuals separate into groups primarily by interest
in local, national or global (cosmopolitan) issues. The national group subdivides into liberal,
conservative and a diverse but mostly business oriented group with sports, arts and other
We thank Karla Bertrand for helpful comments on the manuscript. This work was sup-
ported in part by ONR under grant #N000140910516.
Appendix A: Methodology and Data Collection
We collected tweets that contain a URL from the “nytimes.com” domain, during Septem-
ber 14 - 29, 2011 using the Twitter Application Programming Interface (API). This resulted
in 521,733 tweets posted by 223,950 unique users.
Categories of New York Times online articles are represented in their URLs. For exam-
ple, the URL http://www.nytimes.com/2011/10/05/science/space/05nobel.html points to
an article in the category of science. URL compression is commonly used and is accounted
for in the analysis. As of September 2011, New York Times had more than 50 main cate-
gories including world, us, politics, magazine and so on. We only focused on the 10 most
popular main categories in our sample (Table I).
Category Number of URL links
TABLE I: The New York Times online article categories and the number of URLs belonging to
these categories, in the full sample.
We limited further analysis to users who posted at least three URLs from the same
category. We gathered the “following” relationships (who follows whom) among these users
via the Twitter API, and constructed a graph representing the users and the follow relations
between them. The giant component of the resulting graph contains 8,106 nodes and 163,850
links. Each user was labeled with the category that s/he posted the most. Ties were set to
the less popular categories according to the number of posts in Table I.
Appendix B: Layout generation and clustering
The spatial layout of the network given in Figure 2 was determined solely by the topology
of the network. We used a force-directed layout algorithm, which heuristically optimizes the
layout so that nearby unconnected nodes push each other apart and edges pull the connected
nodes closer . We employed the implementation provided in the network visualization
toolkit Gephi, with edge cutting parameter set to 0.65 and other parameters at their default
We applied the k-means unsupervised clustering algorithm on the two-dimensional coor-
dinates of the nodes to divide users into two communities. For given k, k-means heuristically
identiﬁes k cluster centers such that total within cluster variance is minimal . Using k = 2
we obtained two clusters, A and B, which consisted of 5507 and 2599 users, respectively. To
further reﬁne the clusters, we repeated the separation of each cluster into two sub-clusters
using the same process.
Appendix C: Analysis of geographical locations
The public proﬁle of a Twitter user contains an optional location attribute as an unstruc-
tured text ﬁeld. We analyzed the text in this ﬁeld to characterize where users are from.
References to the same location vary among users. For example, all of the following phrases
are references to New York City: “New York City,” “NYC,” “the big apple,” “Brooklyn,
NY,” “NY City,” “northern manhattan.” To overcome this problem we manually assigned
each of the users into three mutually exclusive categories: New York, US (not including New
York), or international (not including the US). Ambiguous cases were labeled as “other” with
the exception that location ﬁelds which contain only a reference to the United States were
labeled as US. For example, “United States of America” was labeled as the US, “Northern
Europe” was labeled as international, but a single ﬁeld containing the text “California, New
York, Paris” was labeled as “other.”
In Figure 5, the location distribution of each cluster is given.
FIG. 5: The network structure layout given in Figure 2 with edges hidden and users colored
according to their geographical locations. For each cluster, the distribution of the locations are
given in bar charts. The width of each bar-chart component is proportional to the number of
users in the corresponding category. The total width of each bar chart is proportional to the total
number of users in each cluster.
Appendix D: Analysis of overrepresented words
The public proﬁle of a Twitter user contains a short biography (bio) of the user as an
optional, free-text ﬁeld. We analyzed the words occurring in this ﬁeld to characterize how
they describe themselves. Even though these ﬁelds may contain poorly characterized terms
at the individual level (such as non-standard language or typos), it is possible to capture the
main characteristics of communities via aggregating text-based statistics over large number
We employed a bag-of-words approach, a standard text representation technique which
discards the word order and represents a chunk of text as a set of words occurring in the
text . We converted upper to lower-case, used the space character to split the text into
words and did not carry out any lemmatization – the surface forms of the words such as
“works” and “working” were counted separately. We removed any word which contained
a character outside the English alphabet and two accented characters ´e, `e, and -. We also
discarded commonly used words (called stop words) such as “and,” “the” and “of.”
We deﬁne the prevalence of a location term w in a cluster c as the ratio of users in c who
mention w in their location ﬁelds, and denote it by p
(w, c). We denote the analogous
measure for the terms mentioned in the bio ﬁeld by p
(w, c), and for the article topics by
(w, c). For simplicity, when we talk about all three measures, we refer to them simply
In Table II, we list the top-ﬁve words with the highest p
values for the four sub-
clusters. The lists give some idea about how the users in each sub-cluster identify themselves.
However, the rankings of words according to the p measures are dominated by the raw
frequency and it is hard to compare diﬀerent groups of people based on this raw frequency
values. For instance some frequently used words such as “com” and “news” are found in
more than one list. To compare clusters, we use the diﬀerence in prevalences of the terms
in the two clusters. Given two clusters c
, the bias of a word w, is b
) − p
). The sign of this measure is positive for c
cluster dominant words.
Rank Upper A Lower A Upper B Lower B
1 news news politics news
2 world com liberal love
3 human media political business
4 rights writer news marketing
5 politics new-york progressive com
TABLE II: Most prevalent words in the description ﬁelds of users for each of four sub-clusters.
To generate Figure 4, for each attribute (topic, location, and biography), we identiﬁed
the 10 most prevalent terms in each of the four sub-clusters, and computed their bias. The
most A-skewed terms are given on the left and the most B-skewed terms are given on the
right, in decreasing bias order. The bar widths are proportional to the prevalence values of
the terms. For instance, the term “liberal” is among the most dominant biography terms
in cluster B; therefore it is given on the right side of Figure 1. Moreover we also see that
the prevalence of “liberal” in Upper B is much higher than its prevalence in lower B by
comparing the width of the dark orange bar to the width of the light orange bar (which is
The bottom left and right parts of Figure 4 are generated similarly, considering only lower
and upper A (B) respectively.
 J. U. Henrikson, The growth of social media: An infographic, Search
Engine Journal (8/30/2011 http://www.searchenginejournal.com/
 J. B. Thompson, The media and modernity: A social theory of the media (Stanford University
Press, Stanford, CA, 1995).
 A. Briggs, P. Burke, Social History of the Media: From Gutenberg to the Internet (Polity
Press, Cambridge, UK, 2002).
 M. McPherson, L. Smith-Lovin, J. M. Cook, Birds of a feather: Homophily in social networks,
Annual Review of Sociology, 27, 415-444 (2001).
 C. R. Sunstein, Infotopia: How Many Minds Produce Knowledge.
 E. Gilbert, T. Bergstrom, K. Karahalios, Blogs are echo chambers: Blogs are echo chambers,
Proceedings of Hawaii International Conference on System Sciences 2009 (2009).
 M. S. Granovetter, The strength of weak ties, American Journal of Sociology, 78, 1360–1380
 H. A. Simon, The Sciences of the Artiﬁcial, 3rd ed. (MIT Press, Cambridge, MA, 1999).
 Y. Bar-Yam, Dynamics of Complex Systems (Westview Press, Boulder, CO, 1997).
 M. T. Poe, A History of Communications: Media and Society from the Evolution of Speech to
the Internet (Cambridge University Press, New York, NY, 2011).
 B. Winston, Media Technology and Society: A History from the Telegraph to the Internet
(Routledge, New York, 2000).
 Y. Bar-Yam, Twitter rising: Network scientists say social media are just what
our society needs, PhysOrg (March 29, 2011 http://www.physorg.com/news/
 J. Howe, Crowdsourcing: Why the Power of the Crowd is Driving the Future of Business
(Crown Business, New York, NY, 2008).
 J. Surowiecki, The Wisdom of Crowds (Anchor Books, New York, 2005).
 S. Wu, J. M. Hofman, W. A. Mason, D. J. Watts, Who says what to whom on Twitter,
Proceedings of the 20th International Conference on World Wide Web - WWW ’11 705–714
 M. Mendoza, B. Poblete, C. Castillo, Twitter under crisis: Can we trust what we RT?,
Proceedings of the First Workshop on Social Media Analytics 71–79 (2010).
 L. Yu, S. Asur, B. A. Huberman, What Trends in Chinese social media, The 5th SNA-KDD
Workshop ’11 (2011).
 Twitter, Numbers (2011). http://blog.twitter.com/2011/03/numbers.html Fetched Oc-
tober 22, 2011.
 J. Akshay, S. Xiaodan, F. Tim, T. Belle, Why we Twitter: Understanding microblogging
usage and communities, Procedings of the Joint 9th WEBKDD and 1st SNA-KDD Workshop
 M. Naaman, J. Boase, C. Lai, Is it really about me?: Message content in social awareness
streams, Proceedings of the 2010 ACM Conference on Computer Supported Cooperative Work
 B. Suh, L. Hong, P. Pirolli, E. Chi, Want to be retweeted? Large scale analytics on factors
impacting retweet in Twitter network, 2010 IEEE Second International Conference on Social
Computing (SocialCom) 177–184 (2010).
 A. Lenhart, Twitter and status updating: Demographics, mobile access and news con-
sumption, Pew Internet and American Life Project (2009). http://www.pewinternet.org/
Reports/2009/Twitter-and-status-updating.aspx Fetched October 22, 2011.
 H. Kwak, C. Lee, H. Park, S. Moon, What is Twitter, a social network or a news media?,
WWW ’10: Proceedings of the 19th International Conference on World Wide Web 591–600
 M. Newman, The structure and function of complex networks, SIAM Review 167–256 (2003).
 Y. Bar-Yam, I. Epstein, Response of complex networks to stimuli, PNAS, 101, 4341–5 (2004).
 V. Colizza, A. Barrat, M. Barth´elemy, A. Vespignani, The role of the airline transportation
network in the prediction and predictability of global epidemics, PNAS, 103, 2015 (2006).
 K. Lerman, R. Ghosh, Information contagion: An empirical study of the spread of news on
Digg and Twitter social networks, Proceedings of 4th International Conference on Weblogs
and Social Media (ICWSM) (2010).
 S. Martin, W. M. Brown, R. Klavans, K. W. Boyack, Openord: An open-source toolbox for
large graph layout, Conference on Visualization and Data Analysis, 7868, 786806 (2011).
 B. Bishop, The big sort: Why the clustering of like-minded America Is tearing us apart
(Houghton Miﬄin Harcourt, Boston, MA, 2008).
 M. S. Levendusky, The Partisan Sort: How Liberals Became Democrats and Conservatives
Became Republicans (University of Chicago Press, Chicago, IL, 2009).
 B. A. Nardi, D. J. Schiano, M. Gumbrecht, L. Swartz, Why we blog, Commununications of
the ACM, 47, 41–46 (2004).
 D. Zhao, M. B. Rosson, How and why people twitter: the role that micro-blogging plays in
informal communication at work, Proceedings of the ACM 2009 International Conference on
Supporting Group Work 243–252 (2009).
 M. Cha, H. Haddadi, F. Benevenuto, K. P. Gummadi, Measuring user inﬂuence in Twitter:
the million follower fallacy, Proceedings of International AAAI Conference on Weblogs and
Social Media (2010).
 M. Michelson, S. A. Macskassy, Discovering users’ topics of interest on Twitter: A ﬁrst look,
Proceedings of the Fourth Workshop on Analytics for Noisy Unstructured Text Data 73–80
 S. A. Macskassy, M. Michelson, Why do people retweet? Anti-homophily wins the day!,
Proceedings of the Fifth International AAAI Conference on Weblogs and Social Media (2011).
 J. Onnela, S. Arbesman, M. Gonz´alez, A. Barab´asi, N. Christakis, Geographic constraints on
social network groups, PLoS one, 6, e16939 (2011).
 D. Liben-Nowell, J. Novak, R. Kumar, P. Raghavan, A. Tomkins, Geographic routing in social
networks, Proceedings of the National Academy of Sciences of the United States of America,
102, 11623 (2005).
 J. Travers, S. Milgram, An experimental study of the small world problem, Sociometry, 32,
 D. J. Watts, P. S. Dodds, M. E. J. Newman, Identity and search in social networks, Science,
296, 1302-1305 (2002).
 G. Lotan, M. Ananny, D. Gaﬀney, D. Boyd, The revolutions were tweeted: Information ﬂows
during the 2011 Tunisian and Egyptian revolutions, International Journal of Communication,
5, 1375–1405 (2011).
 M. Bastian, S. Heymann, M. Jacomy, Gephi: An open source software for exploring and
manipulating networks, International AAAI Conference on Weblogs and Social Media (2009).
 T. Hastie, R. Tibshirani, J. Friedman, J. Franklin, The elements of statistical learning: Data
mining, inference and prediction (Springer Verlag, 2011).
 A. Kao, S. Poteet, Natural language processing and text mining (Springer Verlag, 2007).